Flow Next Features

/flow-next:features - compounding user-POV drive knowledge

A committed directory (.flow/features/, beside .flow/memory/) records, from the user's point of view, what each user-facing feature is, how a user reaches it, how an agent drives it, and which traps waste a verification run. QA and drive stop re-deriving navigation; work can reuse it to verify what it built.

Split that keeps existing contracts intact: map = how a user gets there (compounds). Spec = what to prove this time. Live drive = proof.

Seed-mode phases live in seed.md; read it only once mode detection resolves MODE=seed. Feature file shape: references/feature-entry-contract.md. Doctor + proof: references/doctor-and-proof.md. Maintain (when MODE=maintain) executes maintain.md.

The skill validates the four-H2 shape itself. Its flowctl calls on the maintain path are features status (read-only: last-proven age and open drift notes), memory list for the feature-map-drift tag handoff, memory mark-stale to retire drift notes whose route proved, and memory add (overlap-judged fold) for bug filing.

This skill and work's feature-map update step (entries its own change altered, feature-map-update.md) are the only map writers; every other stage reads the map and files drift notes (references/feature-entry-contract.md, "Writers and drift notes").

Read working-rules.md first unless you already have this run; it holds for every step of this skill.

Preamble

CRITICAL: flowctl is BUNDLED - NOT installed globally. which flowctl will fail (expected). Define once; subsequent blocks (here and in maintain.md) use $FLOWCTL:

FLOWCTL="${CODEX_HOME:-$HOME/.codex}/scripts/flowctl"
[ -x "$FLOWCTL" ] || FLOWCTL="<plugin-root>/scripts/flowctl"   # <plugin-root> = the directory two levels above this skill's SKILL.md file (the harness gave you that file's absolute path when the skill loaded); substitute it literally
[ -x "$FLOWCTL" ] || FLOWCTL=".flow/bin/flowctl"

Inline skill (no context: fork) - plain-text numbered prompt must stay reachable for the few facts seed cannot observe. Subagents cannot call plain-text numbered prompts (Claude Code issues #12890, #34592). On portable hosts without plain-text numbered prompt, fall back to a plain-text numbered prompt with a final Other - type your own answer option. For read-only scouts use spawn_agent with agent_type: explorer (or the host's generic read-only dispatch with Edit/Write disallowed when Explore is unavailable).

Autonomy refusal

/flow-next:features is user-invoked (or a host loop the human started). flow --auto, land, and every other autonomous driver must not dispatch it. Scan the autonomy-marker namespace, never a fixed two-var list. Any hit refuses with the typed one-line report and stops.

REFUSE=0
# Namespace scan over autonomy marker families. Never a fixed two-var list.
# Families: any FLOW_*AUTONOM* name, plus the mode:autonomous argument token.
if env | grep -E '^FLOW_[^=]*AUTONOM' >/dev/null 2>&1; then
  REFUSE=1
fi
case " ${ARGUMENTS:-} " in
  *" mode:autonomous "*) REFUSE=1 ;;
esac
if [ "$REFUSE" = "1" ]; then
  echo 'FEATURES_VERDICT=REFUSED features=0 reason="autonomy marker present; /flow-next:features is user-invoked only"'
  exit 2
fi

Mode Detection

State-resolved, cwd-relative to the repo root (.flow/features/ under the current working directory). No .flow/features/ directory, or explicit init intent (--init, init, mode:seed, --seed in $ARGUMENTS), routes to seed. A present map with no init intent routes to maintain. Consumers elsewhere gate on directory existence only: no config key, no registration.

MODE="maintain"
if [ ! -d ".flow/features" ]; then
  MODE="seed"
fi
RAW_ARGS="${ARGUMENTS:-}"
for ARG in $RAW_ARGS; do
  case "$ARG" in
    --init|init|mode:seed|--seed) MODE="seed" ;;
  esac
done
printf 'MODE=%s\n' "$MODE"

Interaction Principles

• Interview the repo, not the user. Surface, run command, drive mechanism, observable evidence, isolation: read them from the checkout. Ask only what cannot be observed. Ask the user via plain text. Render the options below as a numbered list 1. … N., followed by a final option N+1. Other — type your own answer. Print the question, then the numbered list, then stop and wait for the user's next message before continuing. Parse the reply as: a bare number 1–N+1 → that option; the literal text of an option label → that option; free text after Other → custom answer.

• Ask via plain-text numbered prompt. Never silently skip the question.

• Prefer multiple choice when natural options exist. Lead with the recommended option and a one-sentence rationale.

• Do not ask before evidence is gathered. Observation first, questions second.

• Multi-surface repos (web + CLI) seed per-surface feature groups under one index. Enumeration is observation: each feature file carries **Surface:**; consumers select by surface + sub-feature IDs.